WIP

.github/workflows/ci.yaml

@@ -92,10 +92,22 @@ jobs:
         with:
           fetch-depth: 0
 
-      - uses: lsst-sqre/build-and-push-to-ghcr@v1
+      - uses: lsst-sqre/build-and-push-to-ghcr@tickets/DM-48192
         id: build-service
         with:
           dockerfile: docker/Dockerfile
+          target: cmservice
+          image: ${{ github.repository }}
+          github_token: ${{ secrets.GITHUB_TOKEN }}
+
+      - uses: lsst-sqre/build-and-push-to-ghcr@tickets/DM-48192
+        id: build-worker
+        with:
+          dockerfile: docker/Dockerfile
+          target: cmworker
+          build-args: |
+            PYTHON_VERSION=3.11
+            UV_VERSION=0.5
           image: ${{ github.repository }}
           github_token: ${{ secrets.GITHUB_TOKEN }}
 

docker-compose.yaml

@@ -22,6 +22,8 @@ services:
       CM_DATABASE_ECHO: true
       CM_DATABASE_URL: postgresql://cm-service@postgresql:5432/cm-service
       CM_DATABASE_PASSWORD: INSECURE-PASSWORD
+    networks:
+      - cmservice
     depends_on:
       postgresql:
         condition: service_healthy
@@ -33,6 +35,7 @@ services:
     build:
       context: .
       dockerfile: docker/Dockerfile
+      target: cmservice
     entrypoint:
       - uvicorn
     command:
@@ -44,6 +47,8 @@ services:
     environment: *cmenv
     ports:
       - "8080:8080"
+    networks:
+      - cmservice
     depends_on:
       init-db:
         condition: service_completed_successfully
@@ -55,12 +60,15 @@ services:
     build:
       context: .
       dockerfile: docker/Dockerfile
+      target: cmworker
     entrypoint:
       - /opt/venv/bin/python3
       - -m
     command:
       - lsst.cmservice.daemon
     environment: *cmenv
+    networks:
+      - cmservice
     depends_on:
       init-db:
         condition: service_completed_successfully
@@ -74,6 +82,8 @@ services:
       POSTGRES_DB: "cm-service"
     ports:
       - "5432"
+    networks:
+      - cmservice
     volumes:
       - "pgsql:/var/lib/postgresql/data"
     healthcheck:
@@ -87,3 +97,7 @@ services:
 
 volumes:
   pgsql:
+
+networks:
+  cmservice:
+    driver: bridge

docker/Dockerfile

@@ -1,13 +1,18 @@
 # syntax=docker/dockerfile:1
 
-ARG PYTHON_VERSION=3.11
-ARG UV_VERSION=0.5
+ARG PYTHON_VERSION
+ARG UV_VERSION
 
 #==============================================================================
 # UV SOURCE IMAGE
 FROM ghcr.io/astral-sh/uv:${UV_VERSION} AS uv
 
 
+# HTCONDOR SOURCE IMAGE
+FROM continuumio/miniconda3:latest AS htcondor
+RUN conda create -p /opt/htcondor -c conda-forge htcondor-utils --no-default-packages
+
+
 #==============================================================================
 # BASE IMAGE
 FROM python:${PYTHON_VERSION}-slim-bookworm AS base-image
@@ -69,8 +74,8 @@ ENDRUN
 
 
 #==============================================================================
-# RUNTIME IMAGE
-FROM base-image AS runtime-image
+# RUNTIME IMAGE - CM-Service
+FROM base-image AS cmservice
 
 # Expose the API port
 EXPOSE 8080
@@ -105,5 +110,12 @@ EOF
 ENTRYPOINT ["./docker-entrypoint.sh"]
 CMD ["lsst.cmservice.main:app", "--host", "0.0.0.0", "--port", "8080"]
 
-# ENTRYPOINT ["/opt/venv/bin/python3", "-m"]
-# CMD ["lsst.cmservice.daemon"]
+
+#==============================================================================
+# RUNTIME IMAGE - Daemon
+FROM cmservice AS cmworker
+
+COPY --from=htcondor /opt/htcondor /opt/htcondor
+
+ENTRYPOINT ["/opt/venv/bin/python3", "-m"]
+CMD ["lsst.cmservice.daemon"]

src/lsst/cmservice/config.py

@@ -1,9 +1,12 @@
+from dotenv import load_dotenv
 from pydantic import Field
 from pydantic_settings import BaseSettings
 from safir.logging import LogLevel, Profile
 
 __all__ = ["Configuration", "config"]
 
+load_dotenv()
+
 
 class Configuration(BaseSettings):
     """Configuration for cm-service."""

src/lsst/cmservice/daemon.py

@@ -1,14 +1,68 @@
-import asyncio
+from asyncio import create_task, sleep
+from collections.abc import AsyncGenerator
+from contextlib import asynccontextmanager
 from datetime import datetime, timedelta
-from time import sleep
 
 import structlog
+import uvicorn
+from fastapi import APIRouter, FastAPI, HTTPException
 from safir.database import create_async_session, create_database_engine
 
+from . import __version__
 from .common.daemon import daemon_iteration
 from .config import config
 
 
+class State:
+    """A class defining an application State object.
+
+    Attributes
+    ----------
+    tasks : set
+        Set of coroutine tasks that have been added to the event loop.
+    """
+
+    tasks: set
+
+    def __init__(self) -> None:
+        self.tasks = set()
+
+
+health_router = APIRouter()
+"""An API Router for a health endpoint"""
+
+state = State()
+"""A global State object"""
+
+
+@health_router.get("/healthz", tags=["internal", "health"])
+async def get_healthz() -> dict:
+    server_ok = True
+    health_response = {}
+
+    for task in state.tasks:
+        task_response: dict[str, bool | str | None] = {"task_running": True, "task_exception": None}
+        if task.done():
+            server_ok = False
+            task_response["task_running"] = False
+            task_response["task_exception"] = str(task.exception())
+        health_response[task.get_name()] = task_response
+
+    if not server_ok:
+        raise HTTPException(status_code=500, detail=health_response)
+    else:
+        return health_response
+
+
+@asynccontextmanager
+async def lifespan(_: FastAPI) -> AsyncGenerator:
+    # start
+    daemon = create_task(main_loop(), name="daemon")
+    state.tasks.add(daemon)
+    yield
+    # stop
+
+
 async def main_loop() -> None:
     logger = structlog.get_logger(config.logger_name)
     engine = create_database_engine(config.database_url, config.database_password)
@@ -26,12 +80,18 @@ async def main_loop() -> None:
                 logger.info(f"Daemon completed {iteration_count} iterations in {delta_seconds} seconds.")
                 last_log_time = datetime.now()
             await daemon_iteration(session)
-            sleep(15)
+            await sleep(15)
 
 
 def main() -> None:
-    loop = asyncio.get_event_loop()
-    loop.run_until_complete(main_loop())
+    app = FastAPI(
+        lifespan=lifespan,
+        version=__version__,
+    )
+
+    app.include_router(health_router)
+
+    uvicorn.run(app, host="0.0.0.0", port=8081)
 
 
 if __name__ == "__main__":